MODEL OBSERVABILITY

模型评测看板

一眼比较各厂商免费模型的首字耗时与生成吞吐，亦可下钻趋势。

最快首字 · 最快延迟1hr · —

58ms

qwen/qwen3.8-27b · Groq Cloud

最快首字 · 111 可用模型

已接入 Provider

17 运行正常

17 / 22

可用模型

111

上次刷新 · 4 分钟前

上次：4 分钟前

厂商概览?

厂商免费额度模型数量注册方式接入指南

Groq Cloud

每天 1,000 次请求

7 个模型

NVIDIA NIM

每分钟 40 次请求

82 个模型

Experiential Labs

免费及活动规则因情况而异

已探测 12 个

共 20 个模型

配置导出

首字最快吞吐最高E2E 最快

完整评测数据真实数据

排名模型厂商

01openai/gpt-oss-120b128K

Groq Cloud359ms237t/s388ms92.1

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "openai/gpt-oss-120b",
  "provider": "Groq Cloud",
  "ttftMs": 359,
  "latencyMs": 388,
  "tokensPerSec": 237.11340206185565,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 90,
    "completionTokens": 92,
    "totalTokens": 182
  },
  "prompt": "Reply with exactly: pong  [seed:4ac41b8768384bc5]",
  "content": "pong  [seed:4ac41b8768384bc5]",
  "checkedAt": "2026-09-10T16:04:23.309Z"
}

指标计算公式付费

TTFT = 359 ms
E2E = 388 ms
Provider 返回 completion_tokens 92 ÷ 端到端耗时 388 ms = 237 t/s
综合分 RRF = 92.1（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/10 16:04:23

02gemini-3.5-flash-lite-free1.05M

AIHubMix634ms117t/s818ms84.6

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "gemini-3.5-flash-lite-free",
  "provider": "AIHubMix",
  "ttftMs": 634,
  "latencyMs": 818,
  "tokensPerSec": 117.359413202934,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 42,
    "completionTokens": 96,
    "totalTokens": 138
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:ed2984846e5b408c]",
  "content": "pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong",
  "checkedAt": "2026-09-18T00:07:19.372Z"
}

指标计算公式费用未知

TTFT = 634 ms
E2E = 818 ms
Provider 返回 completion_tokens 96 ÷ 端到端耗时 818 ms = 117 t/s
综合分 RRF = 84.6（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/18 00:07:19

03qwen/qwen3.8-27b131K

Groq Cloud58ms33.3t/s60ms78.8

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "qwen/qwen3.8-27b",
  "provider": "Groq Cloud",
  "ttftMs": 58,
  "latencyMs": 60,
  "tokensPerSec": 33.333333333333336,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 38,
    "completionTokens": 2,
    "totalTokens": 40
  },
  "prompt": "Reply with exactly: pong  [seed:28f99a9f861c47e9]",
  "content": "pong",
  "checkedAt": "2026-09-10T16:04:55.394Z"
}

指标计算公式付费

TTFT = 58 ms
E2E = 60 ms
Provider 返回 completion_tokens 2 ÷ 端到端耗时 60 ms = 33.3 t/s
综合分 RRF = 78.8（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/10 16:04:55

04deepseek-v4-flash-latest1M

Experiential Labs1,194ms84.6t/s1,230ms76.3

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "deepseek-v4-flash-latest",
  "provider": "Experiential Labs",
  "ttftMs": 1194,
  "latencyMs": 1230,
  "tokensPerSec": 84.55284552845528,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 25,
    "completionTokens": 104,
    "totalTokens": 129
  },
  "prompt": "Reply with exactly: pong  [seed:9ca2343dc9e64bde]",
  "content": "pong",
  "checkedAt": "2026-09-08T22:02:21.833Z"
}

指标计算公式费用未知

TTFT = 1,194 ms
E2E = 1,230 ms
Provider 返回 completion_tokens 104 ÷ 端到端耗时 1,230 ms = 84.6 t/s
综合分 RRF = 76.3（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/8 22:02:21

05deepseek-v4-flash-vision-exp1M

Experiential Labs1,393ms107t/s1,453ms75.5

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "deepseek-v4-flash-vision-exp",
  "provider": "Experiential Labs",
  "ttftMs": 1393,
  "latencyMs": 1453,
  "tokensPerSec": 106.67584308327598,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 21,
    "completionTokens": 155,
    "totalTokens": 176
  },
  "prompt": "Reply with exactly: pong  [seed:f115aca33613441e]",
  "content": "pong",
  "checkedAt": "2026-09-08T21:32:43.314Z"
}

指标计算公式费用未知

TTFT = 1,393 ms
E2E = 1,453 ms
Provider 返回 completion_tokens 155 ÷ 端到端耗时 1,453 ms = 107 t/s
综合分 RRF = 75.5（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/8 21:32:43

06coding-minimax-m3-free1M

AIHubMix806ms69.6t/s1,437ms75.2

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "coding-minimax-m3-free",
  "provider": "AIHubMix",
  "ttftMs": 806,
  "latencyMs": 1437,
  "tokensPerSec": 69.58942240779402,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 213,
    "completionTokens": 100,
    "totalTokens": 313
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:696fccd2b60a4fb7]",
  "content": "<think>The user wants me to repeat the word \"pong\" separated by single spaces, generating as much as possible until I hit the maximum output token limit. Letme do this.</think>\n\npong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong",
  "checkedAt": "2026-10-14T02:04:37.264Z"
}

指标计算公式费用未知

TTFT = 806 ms
E2E = 1,437 ms
Provider 返回 completion_tokens 100 ÷ 端到端耗时 1,437 ms = 69.6 t/s
综合分 RRF = 75.2（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/10/14 02:04:37

07gemini-3-flash-preview-free1.05M

AIHubMix1,407ms66.9t/s1,436ms71.2

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "gemini-3-flash-preview-free",
  "provider": "AIHubMix",
  "ttftMs": 1407,
  "latencyMs": 1436,
  "tokensPerSec": 66.85236768802228,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 41,
    "completionTokens": 96,
    "totalTokens": 137
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:bb345f8efd12483c]",
  "content": "pong",
  "checkedAt": "2026-09-18T00:06:53.818Z"
}

指标计算公式费用未知

TTFT = 1,407 ms
E2E = 1,436 ms
Provider 返回 completion_tokens 96 ÷ 端到端耗时 1,436 ms = 66.9 t/s
综合分 RRF = 71.2（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/18 00:06:53

08claude-haiku-5.5:free未知

Token Harbor854ms54.7t/s1,828ms70.9

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "claude-haiku-5.5:free",
  "provider": "Token Harbor",
  "ttftMs": 854,
  "latencyMs": 1828,
  "tokensPerSec": 54.70459518599562,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 59,
    "completionTokens": 100,
    "totalTokens": 159
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:715504fc6d9d49fe]",
  "content": "I won't generate an endless repetition like that, since it would just pad output until hitting the token limit without serving any real purpose. But I'm happy to help in other ways. For example, I can repeat \"pong\" a set number of times if you tell me how many you'd like, or I can explain why a repetition prompt like this can cause models to",
  "checkedAt": "2026-10-14T09:32:34.192Z"
}

指标计算公式费用未知

TTFT = 854 ms
E2E = 1,828 ms
Provider 返回 completion_tokens 100 ÷ 端到端耗时 1,828 ms = 54.7 t/s
综合分 RRF = 70.9（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/10/14 09:32:34

09google/diffusiongemma-26b-a4b-it262K

NVIDIA NIM784ms128t/s784ms70.6

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "google/diffusiongemma-26b-a4b-it",
  "provider": "NVIDIA NIM",
  "ttftMs": 784,
  "latencyMs": 784,
  "tokensPerSec": 127.55102040816325,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 54,
    "completionTokens": 100,
    "totalTokens": 154
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:e2eb1ac6021a4e68]",
  "content": "pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong",
  "checkedAt": "2026-10-14T13:01:14.992Z"
}

指标计算公式费用未知

TTFT = 784 ms
E2E = 784 ms
Provider 返回 completion_tokens 100 ÷ 端到端耗时 784 ms = 128 t/s
综合分 RRF = 70.6（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/10/14 13:01:14

10gemini-3.6-flash-free1.05M

AIHubMix1,428ms66.7t/s1,440ms70.3

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "gemini-3.6-flash-free",
  "provider": "AIHubMix",
  "ttftMs": 1428,
  "latencyMs": 1440,
  "tokensPerSec": 66.66666666666667,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 40,
    "completionTokens": 96,
    "totalTokens": 136
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:051e9baca70f4fc5]",
  "content": "a7",
  "checkedAt": "2026-09-18T00:35:32.496Z"
}

指标计算公式费用未知

TTFT = 1,428 ms
E2E = 1,440 ms
Provider 返回 completion_tokens 96 ÷ 端到端耗时 1,440 ms = 66.7 t/s
综合分 RRF = 70.3（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/18 00:35:32

11MiniCPM5-1B131K

AMD Radeon Cloud783ms110t/s783ms70.1

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "MiniCPM5-1B",
  "provider": "AMD Radeon Cloud",
  "ttftMs": 783,
  "latencyMs": 783,
  "tokensPerSec": 109.83397190293742,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 28,
    "completionTokens": 86,
    "totalTokens": 114
  },
  "prompt": "Reply with exactly: pong  [seed:34a71fd964b345cf]",
  "content": "\n\npong [seed:34a71fd964b345cf]",
  "checkedAt": "2026-09-09T05:31:09.813Z"
}

指标计算公式费用未知

TTFT = 783 ms
E2E = 783 ms
Provider 返回 completion_tokens 86 ÷ 端到端耗时 783 ms = 110 t/s
综合分 RRF = 70.1（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/9 05:31:09

12nvidia/nemotron-3-nano-30b-a3b262K

OpenRouter1,199ms206t/s1,240ms69.6

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "nvidia/nemotron-3-nano-30b-a3b",
  "provider": "OpenRouter",
  "ttftMs": 1199,
  "latencyMs": 1240,
  "tokensPerSec": 206.45161290322582,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 42,
    "completionTokens": 256,
    "totalTokens": 298
  },
  "prompt": "Reply with exactly: pong  [seed:94a5f6896ce24f6e]",
  "content": "pong  [seed:94a5",
  "checkedAt": "2026-08-29T16:23:12.543Z"
}

指标计算公式付费

TTFT = 1,199 ms
E2E = 1,240 ms
Provider 返回 completion_tokens 256 ÷ 端到端耗时 1,240 ms = 206 t/s
综合分 RRF = 69.6（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/8/29 16:23:12

13ling-3.0-flash-fin-free262K

Experiential Labs1,267ms200t/s1,282ms68.2

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "ling-3.0-flash-fin-free",
  "provider": "Experiential Labs",
  "ttftMs": 1267,
  "latencyMs": 1282,
  "tokensPerSec": 199.68798751950078,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 46,
    "completionTokens": 256,
    "totalTokens": 302
  },
  "prompt": "Reply with exactly: pong  [seed:870040fbf85a491e]",
  "content": "pong  [seed:87",
  "checkedAt": "2026-09-07T00:02:41.716Z"
}

指标计算公式费用未知

TTFT = 1,267 ms
E2E = 1,282 ms
Provider 返回 completion_tokens 256 ÷ 端到端耗时 1,282 ms = 200 t/s
综合分 RRF = 68.2（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/7 00:02:41

14lfm-2.5-2.6b-free未知

AIHubMix1,255ms161t/s1,333ms67.8

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "lfm-2.5-2.6b-free",
  "provider": "AIHubMix",
  "ttftMs": 1255,
  "latencyMs": 1333,
  "tokensPerSec": 160.54013503375845,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 32,
    "completionTokens": 214,
    "totalTokens": 246
  },
  "prompt": "Reply with exactly: pong  [seed:24c4b778f9b148c9]",
  "content": "pong [seed:24c4b778f9b148c9]",
  "checkedAt": "2026-09-13T01:05:55.017Z"
}

指标计算公式费用未知

TTFT = 1,255 ms
E2E = 1,333 ms
Provider 返回 completion_tokens 214 ÷ 端到端耗时 1,333 ms = 161 t/s
综合分 RRF = 67.8（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/13 01:05:55

15minimax/minimax-m2.5205K

OpenRouter2,536ms84.8t/s2,572ms67.3

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "minimax/minimax-m2.5",
  "provider": "OpenRouter",
  "ttftMs": 2536,
  "latencyMs": 2572,
  "tokensPerSec": 84.75894245723173,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 42,
    "completionTokens": 218,
    "totalTokens": 260
  },
  "prompt": "Reply with exactly: pong  [seed:478cfe6c360f42e5]",
  "content": "pong",
  "checkedAt": "2026-08-29T16:23:02.634Z"
}

指标计算公式付费

TTFT = 2,536 ms
E2E = 2,572 ms
Provider 返回 completion_tokens 218 ÷ 端到端耗时 2,572 ms = 84.8 t/s
综合分 RRF = 67.3（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/8/29 16:23:02

16minimaxai/minimax-m31M

NVIDIA NIM516ms3.7t/s534ms66.7

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "minimaxai/minimax-m3",
  "provider": "NVIDIA NIM",
  "ttftMs": 516,
  "latencyMs": 534,
  "tokensPerSec": 3.745318352059925,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 182,
    "completionTokens": 2,
    "totalTokens": 184
  },
  "prompt": "Reply with exactly: pong  [seed:4177ac366d1340d4]",
  "content": "pong",
  "checkedAt": "2026-09-09T08:32:35.015Z"
}

指标计算公式费用未知

TTFT = 516 ms
E2E = 534 ms
Provider 返回 completion_tokens 2 ÷ 端到端耗时 534 ms = 3.7 t/s
综合分 RRF = 66.7（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/9 08:32:35

17gpt-4.1-nano-free1.05M

AIHubMix1,913ms51.8t/s1,931ms65.2

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "gpt-4.1-nano-free",
  "provider": "AIHubMix",
  "ttftMs": 1913,
  "latencyMs": 1931,
  "tokensPerSec": 51.78663904712584,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 41,
    "completionTokens": 100,
    "totalTokens": 141
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:acd158a39ac040de]",
  "content": "pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong pong",
  "checkedAt": "2026-09-20T00:18:22.524Z"
}

指标计算公式费用未知

TTFT = 1,913 ms
E2E = 1,931 ms
Provider 返回 completion_tokens 100 ÷ 端到端耗时 1,931 ms = 51.8 t/s
综合分 RRF = 65.2（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/20 00:18:22

18nemotron-3-super-120b-a12b-free未知

AIHubMix1,110ms88.5t/s1,130ms65.1

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "nemotron-3-super-120b-a12b-free",
  "provider": "AIHubMix",
  "ttftMs": 1110,
  "latencyMs": 1130,
  "tokensPerSec": 88.49557522123895,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 59,
    "completionTokens": 100,
    "totalTokens": 159
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:17c61aa0983d40b2]",
  "content": "We need to output the word \"pong\" repeated, separated by single spaces, until we reach the maximum output token limit. The system didn't specify token limit, but typical limit is maybe 2048 tokens? However we need to generate as many as possible until hitting limit. Since we don't know exact limit, we can produce a long string. Usually the model will stop when reaching token limit. We can just output many \"pong\" separated by spaces. Let's produce",
  "checkedAt": "2026-09-27T01:07:07.757Z"
}

指标计算公式费用未知

TTFT = 1,110 ms
E2E = 1,130 ms
Provider 返回 completion_tokens 100 ÷ 端到端耗时 1,130 ms = 88.5 t/s
综合分 RRF = 65.1（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/27 01:07:07

19deepseek-v4-flash-07311M

Experiential Labs587ms4.0t/s755ms65.0

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "deepseek-v4-flash-0731",
  "provider": "Experiential Labs",
  "ttftMs": 587,
  "latencyMs": 755,
  "tokensPerSec": 3.9735099337748343,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 26,
    "completionTokens": 3,
    "totalTokens": 29
  },
  "prompt": "Reply with exactly: pong  [seed:a86c2f4fff7e4f30]",
  "content": "pong",
  "checkedAt": "2026-09-09T15:04:06.311Z"
}

指标计算公式费用未知

TTFT = 587 ms
E2E = 755 ms
Provider 返回 completion_tokens 3 ÷ 端到端耗时 755 ms = 4.0 t/s
综合分 RRF = 65.0（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/9/9 15:04:06

20coding-minimax-m2.7-free未知

AIHubMix906ms37.9t/s2,640ms64.8

本次采样证据用于核对吞吐、延迟与 Provider 返回 usage 是否一致

完整 JSON

{
  "model": "coding-minimax-m2.7-free",
  "provider": "AIHubMix",
  "ttftMs": 906,
  "latencyMs": 2640,
  "tokensPerSec": 37.878787878787875,
  "tpsQuality": "measured",
  "tokenUsage": {
    "promptTokens": 79,
    "completionTokens": 100,
    "totalTokens": 179
  },
  "prompt": "Repeat the word \"pong\", separated by single spaces. Keep generating until you reach the maximum output token limit.  [seed:bf2e2d8e18364afc]",
  "content": "<think>\nThe user asks: \"Repeat the word 'pong', separated by single spaces. Keep generating until you reach the maximum output token limit. [seed:bf2e2d8e18364afc]\"\n\nWe have to generate a large number of occurrences of \"pong\". The instruction: The user wants to repeat the word \"pong\" separated by spaces, until we reach the maximum output token limit. So we need to generate a huge block of \"pong pong pong ...\" until the\n</think>\n",
  "checkedAt": "2026-10-14T02:04:38.467Z"
}

指标计算公式费用未知

TTFT = 906 ms
E2E = 2,640 ms
Provider 返回 completion_tokens 100 ÷ 端到端耗时 2,640 ms = 37.9 t/s
综合分 RRF = 64.8（吞吐权重 2×；>300 t/s 降为 0.5×）
检查于: 2026/10/14 02:04:38

⌁正在加载趋势正在读取最近 7 天的历史采样。

SVG MODEL STUDY

鹈鹕骑自行车实测

让不同模型用同一句提示词作画，直观看出构图、细节与 SVG 表达差异。仅展示成功生成的作品。